[74604] in North American Network Operators' Group
MED and community fluctuation
daemon@ATHENA.MIT.EDU (Zhen Wu)
Fri Oct 8 14:40:03 2004
Cc: zwu@cs.uoregon.edu
From: Zhen Wu <zwu@cs.uoregon.edu>
Date: Fri, 8 Oct 2004 11:40:54 -0700
To: nanog@merit.edu
Errors-To: owner-nanog-outgoing@merit.edu
I understand the usage and application of MED and community attributes.
But we watch some interesting policy/attribute fluctuation of MED and
community.
We analyze BGP updates sent out by RouteViews peers. We observed a lot
of AADupType2 instabilities. We define AADupType2 event as: A route is
implicitly withdrawn and replaced
with a duplicate of the original route. The two messages have the same
NEXTHOP and ASPATH attribute, but differ between at least one other
attribute (like MED). Note that we only pair two updates from the same
RouteViews' peer / neighboring AS.
A fine-grained analysis revealed that most AADupType2 instabilities
just change the MED or community attribute (or both), which means that
MED and community values associated with prefixes are changing
dynamically and frequently. We are not claiming it is a common practice
of the Internet, this is just the cases observed from RouteViews peers.
We are thinking of the motivation of doing this? Why the ISPs
configured their network so that the MED values oscillate? Did ISPs
configure MED values dynamically calculated from IGP metrics?
The most strange thing to us is why the community attribute associated
with prefixes also oscillate frequently. We know that community are
used to simply operational complexity: group a set of prefixes to apply
the same policy. It didn't make sense to me to change it dynamically.
Misconfigurations?
Any comments or discussions? Thanks!
Zhen Wu
